This article will guide you to using MMSkeleton quickly and introduce you to testing real-time pose estimation with a camera.

  • MMSkeleton: github.com/open-mmlab/…

The installation

First install MMDetection, MMDetection can be used.

Then install the MMSkeleton,

Enable the Python virtual environment
conda activate open-mmlab

# download MMSkeleton
git clone https://github.com/open-mmlab/mmskeleton.git
cd mmskeleton

# installation MMSkeleton
python setup.py develop

Install NMS op for Person Estimation
cd mmskeleton/ops/nms/
python setup_linux.py develop
cd. /.. /.. /Copy the code

Existing models, video testing

configuration

configs/pose_estimation/pose_demo.yaml:

processor_cfg:
  video_file: resource/data_example/ta_chi.mp4
  detection_cfg:
    model_cfg: ../mmdetection/configs/cascade_rcnn/cascade_rcnn_r50_fpn_1x_coco.py
    checkpoint_file: ../mmdetection/checkpoints/cascade_rcnn_r50_fpn_1x_coco_20200316-3dc56deb.pth
    bbox_thre: 0.8
Copy the code

The selected detection model is as follows:

  • Cascade R-CNN, R-50-FPN, 1x
    • config
    • model

run

# verify that mmskeleton and mmdetection installed correctly
# python mmskl.py pose_demo [--gpus $GPUS]
python mmskl.py pose_demo --gpus 1
Copy the code

The results will be saved to work_dir/pose_demo/ ta_ch.mp4.

Existing model, WebCam test

configuration

configs/apis/pose_estimator.cascade_rcnn+hrnet.yaml:

detection_cfg:
  model_cfg: mmdetection/configs/cascade_rcnn/cascade_rcnn_r50_fpn_1x_coco.py
  checkpoint_file: mmdetection/checkpoints/cascade_rcnn_r50_fpn_1x_coco_20200316-3dc56deb.pth
  bbox_thre: 0.8
estimation_cfg:
  model_cfg: mmskeleton/configs/pose_estimation/hrnet/pose_hrnet_w32_256x192_test.yaml
  checkpoint_file: mmskeleton://pose_estimation/pose_hrnet_w32_256x192
  data_cfg:
    image_size:
      - 192
      - 256
    pixel_std: 200
    image_mean:
      - 0.485
      - 0.456
      - 0.406
    image_std:
      - 0.229
      - 0.224
      - 0.225
    post_process: true
Copy the code

Verify that the detection_CFG ESTIMation_CFG path is correct.

Write the code

Write webcam.py, the main code is as follows:

def main() :
  args = parse_args()

  win_name = args.win_name
  cv.namedWindow(win_name, cv.WINDOW_NORMAL)

  with Camera(args.cam_idx, args.cam_width, args.cam_height, args.cam_fps) as cam:
    cfg = mmcv.Config.fromfile(args.cfg_file)
    detection_cfg = cfg["detection_cfg"]

    print("Loading model ...")
    model = init_pose_estimator(**cfg, device=0)
    print("Loading model done")

    for frame in cam.reads():
      res = inference_pose_estimator(model, frame)

      res_image = pose_demo.render(
          frame, res["joint_preds"], res["person_bbox"],
          detection_cfg.bbox_thre)

      cv.imshow(win_name, res_image)

      key = cv.waitKey(1) & 0xFF
      if key == 27 or key == ord("q") :break

  cv.destroyAllWindows()
Copy the code

run

$ python webcam.py \ --cam_idx 2 --cam_width 640 --cam_height 480 --cam_fps 10 \ --cfg_file configs/apis/pose_estimator.cascade_rcnn+hrnet.yaml Args win_name: webcam cam_idx: 2 cam_width: 640 cam_height: Cam_fps: 10 cfg_file: configs/apis/pose_estimator.cascade_rcnn+hrnet.yaml CAM: 640.0x480.0 10.0 Loading Model... Loading modeldone
Copy the code

Effect,

Camera parameters, you can see the WebCam camera used.

More and more

  • Awesome Human Pose Estimation
  • Awesome Skeleton based Action Recognition

GoCoding personal practice experience sharing, can follow the public account!